Will Abdul El-Sayed win the 2026 US Senate election in Michigan?

TL;DR

My forecast is a 67.6% chance that Abdul El-Sayed is elected. His advantage is small: the latest verified survey gives him a two-point lead (InsiderAdvantage, October 11). Rogers wins 32.4% of my modeled outcomes, chiefly through polling error, turnout differences, or late movement.

Context

This assessment uses information available at the client’s October 15, 2026, 10:44 UTC cutoff. Michigan’s official candidate list confirms El-Sayed as the Democratic nominee and Mike Rogers as the Republican nominee; the general election is scheduled for November 3, 2026 (candidate list, official election calendar).

The forecast concerns the ultimate lawful winner under the client’s certification and court-order rules. Winning the primary does not satisfy those rules. I make no separate directional adjustment for recounts or litigation.

Evidence

The historical backbone is Michigan’s polling-error record, not its reputation as a Democratic Senate state. My reference class contains every Michigan general-election contest for president, governor, and U.S. Senate from 2010 through 2024: 13 races across eight election years. The table compares archived final polling averages with certified results; these are final-average errors, not errors measured nineteen days before voting (official results archive).

All margins below are Democratic minus Republican, in percentage points. Actual margins use all votes cast in the contest, including minor-party votes. Negative error means the polls overstated Democrats. Values are rounded; the calculations use the underlying vote totals.

Across this working comparison, the mean error is about +0.4 points and the sample standard deviation is 4.3 points, calculated from the linked polling records and certified returns. Errors within an election year are correlated, so thirteen contests do not supply thirteen independent tests. The recent Senate misses favor Rogers, but the full record does not support an automatic, permanent Republican correction.

Rogers also has a directly relevant electoral result. In the November 5, 2024 Senate election, he lost to Elissa Slotkin by 19,006 votes out of 5,577,187 votes cast—0.34 percentage point. He is a demonstrated statewide competitor, not an untested challenger (certified Michigan results).

El-Sayed’s primary provides another warning. The final RCP average, covering July 8–29, showed him ahead of Haley Stevens by 10.2 points. The August 4 certified totals gave him a 13,792-vote victory out of 1,537,628 Democratic-primary votes, a margin of 0.90 point (primary polling archive, official returns). That is evidence against excessive confidence in his polling. It is not a defensible nine-point subtraction from general-election polls: the electorate differs, and the available evidence does not isolate turnout error from late movement or undecided allocation.

The current lead survives source checking. RCP displays El-Sayed at 48.4% and Rogers at 45.7%, a 2.7-point advantage, using surveys completed September 24–October 10 (current polling ledger). My calculation uses the following eighteen survey records, all available before the cutoff. Dates are fieldwork dates in 2026; LV means likely voters and RV means registered voters. Margins preserve each survey’s reported denominator rather than forcing every poll into two-party shares.

These records are not equally well verified. Original documents support the central narrow-lead evidence. Wedgewood, Guidant, and Z to A receive explicit discounts because their documentation was less complete; Z to A also has a Democratic sponsorship flag. Mitchell’s original figures give +4.2, not the +5 obtained by subtracting separately rounded candidate shares (Mitchell release).

The broader environment supports a Democratic advantage, but El-Sayed does not capture all of it. Marist’s September 24–27 survey found a nine-point Democratic congressional-ballot lead, Trump approval of 41% against 57% disapproval, and a twelve-point Benson gubernatorial lead—alongside El-Sayed’s seven-point Senate lead. These are observations from the same 1,235-RV survey, not independent bonuses to add to his Senate polling (Marist, released September 30).

Republicans have the resources to contest the finish. AdImpact figures reported October 11 put Republican advertising spending through October 10 at $78 million, against $63 million for Democrats (AP). The later $181 million headline includes future advertising reservations, so it should not be read as completed spending or added to those totals (Axios, October 15). I treat the spending advantage as a closing-campaign risk, not another numerical penalty on top of polls already exposed to the advertising.

My aggregation weights decay with a fourteen-day half-life from fieldwork midpoint. Sample-size weight is proportional to the square root of sample size, capped at 1,200 respondents. Marist receives a 0.65 population factor; Wedgewood and Guidant receive 0.50 verification factors, and Z to A receives 0.25. Other factors equal one. These are judgmental rules, not fitted pollster grades.

That calculation gives a polling margin of El-Sayed +2.98946 points. I subtract 0.8 point as a modest allowance for the recent Senate record of Democratic overstatement, while shrinking strongly away from assuming those misses recur in full. This is a judgmental adjustment, not the mean of the complete historical sample. It yields an expected final margin of +2.18946 points.

I use a 4.8-point forecast standard deviation. The historical final-average error spread is about 4.3 points; the larger forecast allowance retains room for remaining movement and aggregation uncertainty. It is not the sampling error of eighteen pooled surveys, and it does not shrink away as more polls repeat the same turnout assumptions.

$$M\sim\mathcal{N}(2.18946,4.8^2),\qquad P(\mathrm{YES})\approx P(M>0)=\Phi(2.18946/4.8)\approx67.6\%.$$

Here, $M$ is El-Sayed’s eventual margin over Rogers in percentage points, and $\Phi$ is the standard normal cumulative distribution function. The central 80% margin interval runs from Rogers +4.0 to El-Sayed +8.3. Minor-party candidates remain far behind in the ballot tests, which supports this two-candidate approximation (Trafalgar full ballot).

Sensitivity matters more than the final decimal. Excluding the three less-verified supplemental surveys gives 65.5%. Removing the directional adjustment gives 73%; increasing it to 1.5 points gives 62.2%. These are alternative assumptions applied to the same evidence, not independent forecasts to average.

I calculated the initial poll-based estimate before consulting market quotations. Pre-cutoff observations showed a Kalshi Democratic midpoint of 68.5% at 09:00 UTC; Polymarket’s Democratic and Republican midpoints at 10:00 UTC were 70.5 and 30.5 cents, implying 69.8% Democratic after normalization (timestamped market-price source). Those prices are consistent with this forecast and supply no missing fact that warrants changing it. Historical order-book depth was not consistently available, so I do not construct a liquidity-weighted market blend. Cook’s treatment of Michigan among its toss-up races is another competitiveness check, not a separate numerical observation (September 23 review).

What's non-obvious

One apparent turnout warning does not survive inspection. Mitchell shows Rogers ahead 47–45 among respondents labeled definitely voting. But that category excludes a separate already-voted/absentee group, which favors El-Sayed 66–21. The groups constitute 78% and 11% of the sample respectively; combining their rounded figures produces an El-Sayed lead of about 3.8 points among those two categories, not a Rogers lead. The subgroup bases remain small, but the definite-only result is not evidence for an additional turnout deduction (Mitchell, October 1 survey, pages 3–4).

The same discipline applies to candidate weakness and likely-voter screens. Fox found El-Sayed ahead 51–48 among registered voters but only 50–49 among likely voters. The narrower LV result already incorporates that turnout adjustment. Fox also placed Rogers at 49%, which contradicts treating his mid-forties readings elsewhere as a hard ceiling. I use the narrow LV result without subtracting the same turnout risk again (Fox original release, September 30).

Uncertainties

The largest gap is fresh measurement. The latest verified original survey completed fieldwork on October 10; it followed the October 8 debate, so the debate is not wholly absent from the evidence. It does not establish the effect of subsequent campaign coverage or isolate who benefited from the debate (debate report, post-debate survey). A fresh, transparent multi-mode survey and validated age-specific turnout comparisons would reduce this uncertainty.

Early ballot counts do not close that gap. Local reporting attributed more than 440,000 returned ballots to state figures on October 14, but those counts do not reveal candidate choices (WKZO). Nor could I verify reproducible, cutoff-consistent Michigan win probabilities directly from all the requested outside forecasters, including Silver Bulletin, FiftyPlusOne, and Metaculus; I have not substituted national Senate-control probabilities. The model’s weighting, directional allowance, and distribution remain judgmental. Its decimal output records the calculation, not equivalent precision about the true chance of winning.